Chapter 3

Normal Modes

Systems with several degrees of freedom appear to be much more complicated than the simple
harmonic oscillator. What we will see in this chapter is that this is an illusion. When we look
at it in the right way, we can see the simple oscillators inside the more complicated system.

Preview
In this chapter, we discuss harmonic oscillation in systems with more than one degree of
freedom.

1. We will write down the equations of motion for a system of particles moving under
general linear restoring forces without damping.

2. Next, we introduce matrices and matrix multiplication and show how they can be used
to simplify the description of the equations of motion derived in the previous section.

3. We will then use time translation invariance and find the irreducible solutions to the
equations of motion in matrix form. This will lead to the idea of “normal modes.” We
then show how to put the normal modes together to construct the general solution to
the equations of motion.

4. * We will introduce the idea of “normal coordinates” and show how they can be used
to automate the solution to the initial value problem.

5. * We will discuss damped forced oscillation in systems with many degrees of freedom.

53
54 CHAPTER 3. NORMAL MODES

3.1 More than One Degree of Freedom

In general, the number of degrees of freedom of a system is the number of independent
coordinates required to specify the system’s configuration. The more degrees of freedom the
system has, the larger the number of independent ways that the system can move. The more
possible motions, you might think, the more complicated the system will be to analyze. In
fact, however, using the tools of linear algebra, we will see that we can deal with systems
with many degrees of freedom in a straightforward way.

3.1.1 Two Coupled Oscillators

¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡
¡ ¡ ¡r ¡ ¡ ¡ ¡ ¡r ¡ ¡

... .. ... .
..... ..... ..... ..... ..... .... ............ ........... ........... ........... ...........
.... .... .... .... ...... ...... ...... .. ..
...........................................................................................................
1 2

Figure 3.1: Two pendulums coupled by a spring.

Consider the system of two pendulums shown in figure 3.1. The pendulums consist of
rigid rods pivoted at the top so they oscillate without friction in the plane of the paper. The
masses at the ends of the rods are coupled by a spring. We will consider the free motion of the
system, with no external forces other than gravity. This is a classic example of two “coupled
oscillators.” The spring that connects the two oscillators is the coupling. We will assume that
the spring in figure 3.1 is unstretched when the two pendulums are hanging straight down, as
shown. Then the equilibrium configuration is that shown in figure 3.1. This is an example
of a system with two degrees of freedom, because two quantities, the displacements of each
of the two blocks from equilibrium, are required to specify the configuration of the system.
For example, if the oscillations are small, we can specify the configuration by giving the
horizontal displacement of each of the two blocks from the equilibrium position.
Suppose that block 1 has mass m1 , block 2 has mass m2 , both pendulums have length `
and the spring constant is κ (Greek letter kappa). Label the (small) horizontal displacements
of the blocks to the right, x1 and x2 , as shown in figure 3.2. We could have called these
3.1. MORE THAN ONE DEGREE OF FREEDOM 55

¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡
¡ ¡ ¡..... ¡ ¡ ¡ ¡ ¡..... ¡ ¡
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
...
... ...
... ...
.. .
........
......... ........ ........ ........ ........ ........ .........
.. .. ...
. ...
.. . ...
... .. ....... ........... ...........
.. ... . .. . .. . . . .. .. ........ ....
1 ....................... ............. ............ .......... ......... ..... 2

| | | |
x1 x2
Figure 3.2: Two pendulums coupled by a spring displaced from their equilibrium positions.

masses and displacements anything, but it is very convenient to use the same symbol, x, with
different subscripts. We can then write Newton’s law, F = ma, in a compact and useful
form.
d2
mj 2 xj = Fj , (3.1)
dt
for j = 1 to 2, where F1 is the horizontal force on block 1 and F2 is the horizontal force
on block 2. Because there are two values of j, (3.1) is two equations; one for j = 1 and
another for j = 2. These are the two equations of motion for the system with two degrees
of freedom. We will often refer to all the masses, displacements or forces at once as mj , xj
or Fj , respectively. For example, we will say that Fj is the horizontal force on the jth block.
This is an example of the use of “indices” (j is an index) to simplify the description of a
system with more than one degree of freedom.
When the blocks move horizontally, they will move vertically as well, because the length
of the pendulums remains fixed. Because the vertical displacement is second order in the xj s,

x2j
yj ≈ , (3.2)
2
we can ignore it in thinking about the spring. The spring stays approximately horizontal for
small oscillations.
To find the equation of motion for this system, we must find the forces, Fj , in terms of
the displacements, xj . It is the approximate linearity of the system that allows us to do this
in a useful way. The forces produced by the Hooke’s law spring, and the horizontal forces
on the pendulums due to the tension in the string (which in turn is due to gravity) are both
approximately linear functions of the displacements for small displacements. Furthermore,
the forces vanish when both the displacements vanish, because the system is in equilibrium.
56 CHAPTER 3. NORMAL MODES

Thus each of the forces is some constant (different for each block) times x1 plus some other
constant times x2 . It is convenient to write this as follows:

F1 = −K11 x1 − K12 x2 , F2 = −K21 x1 − K22 x2 , (3.3)

or more compactly,
2
X
Fj = − Kjk xk (3.4)
k=1
for j = 1 to 2. We have written the four constants as K11 , K12 , K21 and K22 in order to
write the force in this compact way. Later, we will call these constants the matrix elements
of the K matrix. In this notation, the equations of motion are

X2
d2
mj 2 xj = − Kjk xk (3.5)
dt k=1

for j = 1 to 2.

¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡
¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡
C
C
C
C
C
C
C
.. ... .... .... ..... ..... ..... .......
C
....... ...... ...... ...... ..... ..... ...... ....
. .
.... ...... ...... ........ ........ .......... ............ .................
1 ....................... ............. ............ .......... ......... ...... ... 2
| |
x2
Figure 3.3: Two pendulums coupled by a spring with block 2 displaced from an equilibrium
position.

Because of the linearity of the system, we can find the constants, Kjk , by considering
the displacements of the blocks one at a time. Then we find the total force using (3.4). For
example, suppose we displace block 2 with block 1 held fixed in its equilibrium position and
look at the forces on both blocks. This will allow us to compute K12 and K22 . The system
with block two displaced is shown in figure 3.3. The forces on the blocks are shown in
figure 3.4, where Tj is the tension in the jth pendulum string. F12 is the force on block 1 due
to the displacement of block 2. F22 is the force on block 2 due to the displacement of block 2.
For small displacements, the restoring force from the spring is nearly horizontal and equal to
 3.1. MORE THAN ONE DEGREE OF FREEDOM 57

κx2 on block 1 and −κx2 on block 2. Likewise, in the limit of small displacement, the vertical
component of the force from the tension T2 nearly cancels the gravitational force on block 2,
m2 g, so that the horizontal component of the tension gives a restoring force −x2 m2 g/` on
block 2. For block 1, the force from the tension T1 just cancels the gravitational force m1 g.
Thus
m2 gx2
F12 ≈ κx2 , F22 ≈ − − κx2 , (3.6)
`
and
m2 g
K12 ≈ −κ , K22 ≈ + κ. (3.7)
`
An analogous argument shows that
m1 g
K21 ≈ −κ , K11 ≈ + κ. (3.8)
`
Notice that
K12 = K21 . (3.9)
We will see below that this is an example of a very general relation.
.....
..... ........
........ ...
...
.... ...
... ...
... ...
... ...
...
... T1
...
...
...
T2
... ...
... ...
... ...
... ...
...
...
.. F12 F22 ...
...

.................................
1 ................................. 2
...
...
...
... κx2 −κx2 ...
...
...
... ...
... ...
... ...
... ...
...
...
... m1 g ...
...
...
m2 g
... ...
... ...
... ...
... ..
.. .........
......... ...
...

Figure 3.4: The forces on the two blocks in figure 3.3.

3.1.2 Linearity and Normal Modes

......................
...... .....
3-1 ...........
......
............ ..............................

We will see in this chapter that the most general possible motion of this system, and of any
such system of oscillators, can be decomposed into particularly simple solutions, in which all
the degrees of freedom oscillate with the same frequency. These simple solutions are called
“normal modes.” The displacements for the most general motion can be written as sums of
the simple solutions. We will study how this works in detail later, but it may be useful to see it
58 CHAPTER 3. NORMAL MODES

first. A possible motion of the system of two coupled oscillators is animated in program 3-1.
Below the actual motion, we show the two simple motions into which the more complicated
motion can be decomposed. For this system, the normal mode with the lower frequency is
one in which the displacements of the two blocks are the same:

x1 (t) = x2 (t) = b1 cos(ω1 t − θ1 ) . (3.10)

The other normal mode is one in which the displacements of the two blocks are opposite

x1 (t) = −x2 (t) = b2 cos(ω2 t − θ2 ) . (3.11)

The sum of these two simple motions gives the much more complicated motion shown in
program 3-1.

3.1.3 n Coupled Oscillators

Before we try to solve the equations of motion, (3.5), let us generalize the discussion to
systems with more degrees of freedom. Consider the oscillation of a system of n particles
connected by various springs with no damping. Our analysis will be completely general, but
for simplicity, we will talk about the particles as if they are constrained to move in the x
direction, so that we can measure the displacement of the jth particle from equilibrium with
the coordinate xj . Then the equilibrium configuration is the one in which all the xj s are all
zero.
Newton’s law, F = ma, for the motion of the system gives

d2 xj
mj = Fj (3.12)
dt2
where mj is the mass of the jth particle, Fj is the force on it. Because the system is linear,
we expect that we can write the force as follows (as in (3.4)):
n
X
Fj = − Kjk xk (3.13)
k=1

for j = 1 to n. The constant, −Kjk , is the force per unit displacement of the jth particle due
to a displacement xk of the kth particle. Note that all the Fj s vanish at equilibrium when all
the xj s are zero. Thus the equations of motion are

d2 xj X
mj 2
=− Kjk xk (3.14)
dt k

for j = 1 to n.
3.2. MATRICES 59

To measure Kjk , make a small displacement, xk , of the kth particle, keeping all the
other particles fixed at zero, assumed to be an equilibrium position. Then measure the
force, Fjk on the jth particle with only the kth particle displaced. Since the system is
linear (because it is made out of springs or in general, as long as the displacement is
small enough), the force is proportional to the displacement, xk . The ratio of Fjk to xk
is −Kjk :
Kjk = −Fjk /xk when x` = 0 for ` = 6 k. (3.15)
Note that Kjk is defined with a − sign, so that a positive K is a force that is opposite to the
displacement, and therefore tends to return the system to equilibrium.
Because the system is linear, the total force due to an arbitrary displacement is the sum
of the contributions from each displacement. Thus
X X
Fj = Fjk = − Kjk xk . (3.16)
k k

Let us now try to understand (3.9). If we consider systems with no damping, the forces
can be derived from a potential energy,
∂V
Fj = − . (3.17)
∂xj
But then by differentiating equation (3.16) we find that

∂2V
Kjk = . (3.18)
∂xj ∂xk
The partial differentiations commute with one another, thus equation (3.18) implies

Kjk = Kkj . (3.19)

In words, the force on particle j due to a displacement of particle k is equal to the force on
particle k due to the displacement of particle j.

3.2 Matrices
It is very useful to rewrite equation (3.14) in a matrix notation. Because of the linearity of
the equations of motion for harmonic motion, it will be very useful to have the tools of linear
algebra at hand for our study of wave phenomena. If you haven’t studied linear algebra (or
didn’t understand much of it) in math courses, DON’T PANIC. We will start from scratch by
describing the properties of matrices and matrix multiplication. The important thing to keep
in mind is that matrices are nothing very deep or magical. They are just bookkeeping devices
designed to make your life easier when you deal with more than one equation at a time.
60 CHAPTER 3. NORMAL MODES

A matrix is a rectangular array of numbers. An N ×M matrix has N rows and M

columns. Matrices can be added and subtracted simply by adding and subtracting each of
the components. The difference comes in multiplication. It is very convenient to define
a multiplication law that defines the product of an N ×M matrix on the left with a M ×L
matrix on the right (the order is important!) to be an N ×L matrix as follows:
Call the N ×M matrix A and let Ajk be the number in the jth row and kth column for
1 ≤ j ≤ N and 1 ≤ k ≤ M . These individual components of the matrix are called matrix
elements. In terms of its matrix elements, the matrix A looks like:
⎛ ⎞
A11 A12 · · · A1M
⎜ A21 A22 · · · A2M ⎟
⎜ ⎟
A = ⎜ .. ... .. ... ⎟ . (3.20)
⎝ . . ⎠
AN 1 AN 2 · · · AN M
Call the M ×L matrix B with matrix elements Bkl for 1 ≤ k ≤ M and 1 ≤ l ≤ L:
⎛ ⎞
B11 B12 · · · B1L
⎜ B21 B22 · · · B2L ⎟
⎜ ⎟
B = ⎜ .. ... .. ... ⎟. (3.21)
⎝ . . ⎠
BM 1 BM 2 · · · BM L
Call the N ×L matrix C with matrix elements Cjl for 1 ≤ j ≤ N and 1 ≤ l ≤ L.
⎛ ⎞
C11 C12 ··· C1L
⎜ C21 C22 ··· C2L ⎟
⎜ ⎟
C=⎜ . . .. .. ⎟. (3.22)
⎝ .. .. . . ⎠
CN 1 CN 2 · · · CN L
Then the matrix C is defined to be the product matrix A B if
M
X
Cjl = Ajk · Bkl . (3.23)
k=1
Equation (3.23) is the algebraic statement of the “row-column” rule. To compute the
j` matrix element of the product matrix, AB, take the jth row of the matrix A and the `th
column of the matrix B and form their dot-product (corresponding to the sum over k in
(3.23)). This rule is illustrated below:
⎛ ⎞⎛ ⎞
⎜ A11 ··· A1k ··· A1M ⎟⎜ B11 ··· B1` ··· B1L ⎟
⎜ .. .. .. ⎟⎜ .. .. ⎟
⎜
⎜ .
..
. .
..
. . ⎟⎜
⎟⎜ .
..
. .
..
. ...⎟
⎟
⎜ ⎟⎜ ⎟
⎜ ⎟⎜ ⎟
⎜ Aj1
⎜ ··· Ajk ··· AjM ⎟ ⎜
⎟ ⎜ Bk1 ··· Bk` ··· BkL ⎟
⎟
⎜ . .. .. .. .. ⎟ ⎜ . .. . .. .. ⎟
⎜ ..
⎜ . . . . ⎟ ⎜ .
⎟⎜ . . .. . . ⎟⎟
⎝ ⎠⎝ ⎠
AN 1 · · · AN k · · · AN M BM 1 · · · BM ` · · · BM L
3.2. MATRICES 61
⎛ ⎞
⎜ C11 ··· C1` ··· C1L ⎟
⎜ .. ⎟
⎜
⎜ .
..
. ... ..
. ...⎟
⎟
⎜ ⎟
⎜ ⎟
= ⎜ Cj1 ··· Cj` ··· CjL ⎟ . (3.24)
⎜ ⎟
⎜ . .. . .. . ⎟
⎜ ..
⎜ . .. . .. ⎟
⎟
⎝ ⎠
CN 1 · · · CN ` · · · CN L
For example, ⎛ ⎞ ⎛ ⎞
2 3 µ ¶ 2 3 13
⎝0 1 0 2
1 ⎠· = ⎝0 1 3 ⎠. (3.25)
0 1 3
2 −1 2 −1 1
It is easy to check that the matrix product defined in this way is associative, (AB)C =
A(BC). However, in general, it is not commutative, AB = 6 BA. In fact, if the matrices
are not square, the product in the opposite order may not even make any sense! The matrix
product AB only makes sense if the number of columns of A is the same as the number of
rows of B. Beware!
Except for the fact that it is not commutative, matrix multiplication behaves very much
like ordinary multiplication. For example, there are “identity” matrices. The N ×N identity
matrix, called I, has zeros everywhere except for 1’s down the diagonal. For example, the
3 × 3 identity matrix is ⎛ ⎞
1 0 0
I = ⎝0 1 0⎠ . (3.26)
0 0 1
The N ×N identity matrix satisfies

IA = AI = A for any N ×N matrix A;

IB = B for any N ×M matrix B; (3.27)

CI = C for any M ×N matrix C.

We will be primarily concerned with “square” (that is N ×N ) matrices.

Matrices allow us to deal with many linear equations at the same time.
An N dimensional column vector can be regarded as an N ×1 matrix. We will call this
object an “N -vector.” It should not be confused with a coordinate vector in three-dimensional
space. Likewise, we can think of an N dimensional row vector as a 1×N matrix. Matrix
multiplication can also describe the product of a matrix with a vector to give a vector. The
62 CHAPTER 3. NORMAL MODES

particularly important case that we will need in order to analyze wave phenomena involves
square matrices. Consider an N ×N matrix A multiplying an N -vector, X, to give another
N -vector, F . The square matrix A has N 2 matrix elements, Ajk for j and k = 1 to N . The
vectors X and F each have N matrix elements, just their components Xj and Fj for j = 1 to
N . Then the matrix equation:
AX = F (3.28)
actually stands for N equations:
N
X
Ajk · Xk = Fj (3.29)
k=1

for j=1 to N . In other words, these are N simultaneous linear equations for the N Xj ’s. You
all know, from your studies of algebra how to solve for the Xj ’s in terms of the Fj ’s and the
Ajk ’s but it is very useful to do it in matrix notation. Sometimes, we can find the “inverse”
of the matrix A, A−1 , which has the property

A A−1 = A−1 A = I , (3.30)

where I is the identity matrix discussed in (3.26) and (3.27). If we can find such a matrix,
then the N simultaneous linear equations, (3.29), have a unique solution that we can write in
a very compact form. Multiply both sides of (3.29) by A−1 . On the left-hand side, we can
use (3.30) and (3.27) to get rid of the A−1 A and write the solution as follows:

X = A−1 F . (3.31)

3.2.1 * Inverse and Determinant

We can compute A−1 in terms of the “determinant” of A. The determinant of the matrix A
is a sum of products of the matrix elements of A with the following properties:

• There are N ! terms in the sum;

• Each term in the sum is a product of N different matrix elements;

• In each product, every row number and every column number appears exactly once;

• Every such product can be obtained from the product of the diagonal elements, A11 A22 · · · AN N ,
by a sequence of interchanges of the column labels. For example, A12 A21 A33 · · · AN N
involves one interchange while A12 A23 A31 A44 · · · AN N requires two.

• The coefficient of a product in the determinant is +1 if it involves an even number of

interchanges and −1 if it involves an odd number of interchanges.
3.2. MATRICES 63

Thus the determinant of a 2×2 matrix, A is

det A = A11 A22 − A12 A21 . (3.32)

The determinant of a 3×3 matrix, A is

det A = A11 A22 A33 + A12 A23 A31 + A13 A21 A32
(3.33)
−A11 A23 A32 − A13 A22 A31 − A12 A21 A33 .

Unless you are very unlucky, you will never have to compute the determinant of a matrix
larger than 3 × 3 by hand. If you are so unlucky, it is best to use an inductive procedure that
builds it up from the determinants of smaller submatrices. We will discuss this procedure
below.
If det A = 0, the matrix has no inverse. It is not “invertible.” In this case, the simul-
taneous linear equations have either no solution at all, or an infinite number of solutions. If
det A 6= 0, the inverse matrix exists and is uniquely given by

A˜
A−1 = (3.34)
det A
where à is the cofactor matrix defined by its matrix elements as follows:

(Ã)jk = det A(jk) (3.35)

with
A(jk)lm = 1 if m = j and l = k;
A(jk)lm = 0 if m = j and l 6= k;
A(jk)lm = 0 if m 6= j and l = k;
A(jk)lm = Alm if m 6= j and l 6= k.
In other words, A(jk) is obtained from the matrix A by replacing the kj matrix element by
1 and all other matrix elements in row k or column j by 0. Thus if
⎛ ⎞
⎜ A11 ··· A1j ··· A1N ⎟
⎜ .. .. .. .. .. ⎟
⎜ . . . . . ⎟
⎜ ⎟
⎜ ⎟
⎜ ⎟
A = ⎜ Ak1 ··· Akj ··· AkN ⎟ , (3.36)
⎜ ⎟
⎜ . .. .. .. ⎟
.. ⎟
⎜ .. . . . . ⎟
⎜
⎝ ⎠
AN 1 · · · AN j · · · AN N
64 CHAPTER 3. NORMAL MODES
⎛ ⎞
⎜ A11 ··· 0 ··· A1N ⎟
⎜ .. . . .. . . .. ⎟
⎜ . . . . . ⎟
⎜ ⎟
⎜ ⎟
⎜ ⎟
A(jk) = ⎜ 0 ··· 1 ··· 0 ⎟. (3.37)
⎜ ⎟
⎜ .. . . .. . . .. ⎟
⎜ . . . . . ⎟
⎜ ⎟
⎝ ⎠
AN 1 · · · 0 · · · AN N
Note the sneaky interchange of j ↔ k in this definition, compared to (3.23).
For example if µ ¶
4 3
A= (3.38)
5 2
then
µ ¶ µ ¶
1 0 0 3
A(11) = A(12) =
0 2 1 0 (3.39)
µ ¶ µ ¶
0 1 4 0
A(21) = A(22) = .
5 0 0 1
Thus, µ ¶
2 −3
à = (3.40)
−5 4
and since det A = 4 · 2 − 5 · 3 = −7,
µ ¶
−2/7 3/7
A−1 = . (3.41)
5/7 −4/7

A−1 satisfies A A−1 = A−1 A = I where I is the identity matrix:

µ ¶
1 0
I= . (3.42)
0 1

In terms of the submatrices, A(jk), we can define the determinant inductively, as promised
above. In fact, the reason that (3.30) works is that the determinant can be written as
N
X
det A = A1k det A(k1) . (3.43)
k=1

Actually this is true for any row, not just j = 1. The relation, (3.30) can be rewritten as
⎧
⎪
⎨
N
X det A for j = j 0
Ajk det A(kj 0 ) = (3.44)
⎪
k=1 ⎩ 0 for j 6= j 0
3.2. MATRICES 65

The determinants of the submatrices, det A(k1), in (3.43) can, in turn, be computed by the
same procedure. The result is a definition of the determinant that refers to itself. However,
eventually, the process terminates because the matrices keep getting smaller and the determi-
nant can always be computed in this way. The only problem with this procedure is that it is
very tedious for a large matrix. For an n × n matrix, you end up computing n! terms and
adding them up. For large n, this is impractical. One of the nice features of the techniques
that we will discuss in the coming chapters is that we will be able to avoid such calculations.

3.2.2 More Useful Facts about Matrices

Suppose that A and B are N ×N matrices and v is an N -vector.
1. If you know the inverses of A and B, you can find the inverse of the product, AB, by
multiplying the inverses in the reverse order:
(A B)−1 = B −1 A−1 . (3.45)

2. The determinant of the product, AB, is the product of the determinants:

det(A B) = det A det B , (3.46)
thus if det(AB) = 0, then either A or B has vanishing determinant.
3. A matrix multiplying a nonzero vector can give zero only if the determinant of the
matrix vanishes:
A v = 0 ⇒ det A = 0 or v = 0 . (3.47)
This is the statement, in matrix language, that N homogeneous linear equations in
N unknowns can have a nontrivial solution, v 6= 0, only if the determinant of the
coefficients vanishes.
4. Similarly, if det A = 0, there exists a nonzero vector, v, that is annihilated by A:
det A = 0 ⇒ ∃ v 6= 0 such that A v = 0 . (3.48)
This is the statement, in matrix language, that N homogeneous linear equations in N
unknowns actually do have a nontrivial solution, v = 6 0, if the determinant of the
coefficients vanishes.
5. The transpose of an N ×M matrix A, denoted by AT , is the M ×N matrix obtained by
reflecting the matrix about a diagonal line through the upper left-hand corner. Thus if
⎛ ⎞
A11 A12 ··· A1M
⎜ A21 A22 ··· A2M ⎟
⎜ ⎟
⎜ . . .. . ⎟
A=⎜
⎜ .. .. . .. ⎟
⎟ (3.49)
⎜ .. .. .. .. ⎟
⎝ . . . . ⎠
AN 1 AN 2 · · · AN M
66 CHAPTER 3. NORMAL MODES

then ⎛ ⎞
A11 A21 ··· ··· AN 1
⎜ A12 A22 ··· ··· AN 2 ⎟
⎜ ⎟
AT = ⎜ .. .. .. .. ... ⎟. (3.50)
⎝ . . . . ⎠
A1M A2M ··· · · · AN M
Note that if N = 6 M , the shape of the matrix is changed by transposition. Only for
square matrices does the transpose give you back a matrix of the same kind. A square
matrix that is equal to its transpose is called a “symmetric” matrix.

3.2.3 Eigenvalue Equations

We will make extensive use of the concept of an “eigenvalue equation.” For an N ×N matrix,
R, the eigenvalue equation has the form:

Rc = hc, (3.51)

where c is a nonzero N -vector,1 and h is a number. The idea is to find both the number, h,
which is called the eigenvalue, and the vector, c, which is called the eigenvector. This is the
problem we discussed in chapter 1 in (1.78) in connection with time translation invariance,
but now written in matrix form.
A couple of examples may be in order. Suppose that R is a diagonal matrix, like
µ ¶
2 0
R= . (3.52)
0 1
Then the eigenvalues are just the diagonal elements, 2 and 1, and the eigenvectors are vectors
in the coordinate directions,
µ ¶ µ ¶ µ ¶ µ ¶
1 1 0 0
R =2 , R =1 . (3.53)
0 0 1 1
A less obvious example is µ ¶
2 1
R= . (3.54)
1 2
This time the eigenvalues are 3 and 1, and the eigenvectors are as shown below:
µ ¶ µ ¶ µ ¶ µ ¶
1 1 1 1
R =3 , R =1 . (3.55)
1 1 −1 −1
It may seem odd that in the eigenvalue equation, both the eigenvalue and the eigenvector
are unknowns. The reason that it works is that for most values of h, the equation, (3.51), has
1
c = 0 doesn’t count, because the equation is satisfied trivially for any h. We are interested only in nontrivial
solutions.
3.2. MATRICES 67

no solution. To see this, we write (3.51) as a set of homogeneous linear equations for the
components of the eigenvector, c,

(R − hI) c = 0 . (3.56)

The set of equations, (3.56), has nonzero solutions for c only if the determinant of the coef-
ficient matrix, R − hI, vanishes. But this will happen only for N values of h, because the
condition
det (R − hI) = 0 (3.57)
is an N th order equation for h. For each h that solves (3.57), we can find a solution for c.2
We will give some examples of this procedure below.

3.2.4 The Matrix Equation of Motion

It is very useful to rewrite the equation of motion, (3.14), in a matrix notation. Define a
column vector, X, whose jth row (from the top) is the coordinate xj :
⎛ ⎞
x1
⎜ x2 ⎟
⎜ ⎟
X = ⎜ .. ⎟ . (3.58)
⎝ . ⎠
xn

Define the “K matrix”, an n×n matrix that has the coefficient Kjk in its jth row and kth
column: ⎛ ⎞
K11 K12 ··· K1n
⎜ K21 K22 ··· K2n ⎟
⎜ ⎟
K = ⎜ .. . .. .. ⎟ . (3.59)
⎝ . .. . . ⎠
Kn1 Kn2 · · · Knn
Kjk is said to be the “jk matrix element” of the K matrix. Because of equation (3.19), the
matrix K is symmetric, K = K T .
Define the diagonal matrix M with mj in the jth row and jth column and zeroes else-
where ⎛ ⎞
m1 0 · · · 0
⎜ 0 m2 · · · 0 ⎟
⎜ ⎟
M = ⎜ .. .. .. . ⎟.
. (3.60)
⎝ . . . . ⎠
0 0 · · · mn
M is called the “mass matrix.”
2
The situation is slightly more complicated when the solutions for h are degenerate. We discuss this in (3.117)
below.
68 CHAPTER 3. NORMAL MODES

Using these definitions, we can rewrite (3.14) in matrix notation as follows:

d2 X
M = −K X . (3.61)
dt2
There is nothing very fancy going on here. We have just used the matrix notation to get rid
of the summation sign in (3.14). The sum is now implicit in the matrix multiplication in
(3.61). This is useful because we can now use the properties of matrices and matrix multi-
plication discussed above to manipulate (3.61). For example, we can simplify (3.61) a bit by
multiplying on the left by M −1 to get
d2 X
= −M −1 K X . (3.62)
dt2

3.3 Normal Modes

If there is only one degree of freedom, then both X and M −1 are just numbers and the
solutions to the equation of motion, (3.62), have the form of a constant amplitude times an
exponential factor. In fact, we saw that this form is related to a very general fact about the
physics – time translation invariance, (1.33). The arguments of chapter 1, (1.71)-(1.85), did
not depend on the number of degrees of freedom. Thus they show that here again, we can
find irreducible solutions, that go into themselves up to an overall constant when the clocks
are reset. As in chapter 1, the first step is to allow the solutions to be complex. That is, we
replace (3.62) by
d2 Z
= −M −1 K Z , (3.63)
dt2
where Z is a complex n vector with components, zj . The real parts of the components of Z
are the components of a real solution satisfying (3.62),
xj = Re zj . (3.64)
We will say that the real vector, X, is the real part of the complex vector, Z,
X = Re Z , (3.65)
if (3.64) is satisfied.
Just as in chapter 1, we know that we can find irreducible solutions that have the same
form up to an overall constant when the clocks are reset. We know from (1.85) that these
have the form
Z(t) = A e−iωt (3.66)
where A is some constant n-vector and the angular frequency, ω, is still just a number. Now
if t → t + a,
Z(t) → Z(t + a) = e−iωa Z(t) . (3.67)
3.3. NORMAL MODES 69

While the irreducible form, (3.66), comes just from time translation invariance, we must
still look at the equations of motion to determine the vector, A and the angular frequency,
ω. Inserting (3.66) into (3.63), doing the differentiation and canceling the exponential factors
from both sides, we find that (3.66) is a solution if

ω 2 A = M −1 K A . (3.68)

This matrix equation is an eigenvalue equation of the form that we discussed in (3.51)-(3.57).
ω 2 is the eigenvalue of the matrix M −1 K and A is the corresponding eigenvector. Let us see
what it means physically.
The real part of the column vector Z specifies the displacement of each of the degrees
of freedom of the system. The eigenvalue equation, (3.68), does not involve any complex
numbers (because we have not put in any damping). Therefore (as we will see explicitly
below), we can choose the solutions so that all the components of A are real. Then the real
part of the complex solutions we seek in (3.66) is

X(t) = A cos ωt , (3.69)

or in terms of the components of A,

⎛ ⎞
a1
⎜a ⎟
A = ⎝ 2⎠ , (3.70)
..
.

x1 (t) = a1 cos ωt , x2 (t) = a2 cos ωt , etc. (3.71)

Not only does everything move with the same frequency, but the ratios of displacements
of the individual degrees of freedom are fixed. Everything oscillates in phase. The only
difference between the motion of the different degrees of freedom is their different amplitudes
from the different components of A.
The point is worth repeating. Time translation invariance and linearity imply that we can
always find irreducible solutions, (3.67), in which all the degrees of freedom oscillate with
the same frequency. The extra piece of information that leads to (3.69) is dynamical. If there
is no damping, then all the components of A can be chosen to be real, and all the degrees of
freedom oscillate not only with the same frequency, but also with the same phase.
If such a solution is to satisfy the equations of motion, then the acceleration must also
be proportional to A, so that the individual displacements don’t get out of synch. But that is
what (3.68) is telling us. −M −1 K is the matrix that, acting on the displacement, gives the
acceleration. The eigenvalue equation (3.68) means that the acceleration is proportional to A
again. The constant of proportionality, ω 2 , is the return force per unit displacement per unit
mass for the particular displacement specified by A.
70 CHAPTER 3. NORMAL MODES

We have already discussed the mathematical structure of the eigenvalue equation in

(3.51)-(3.57). We will do it again, for emphasis, in the case of physical interest, (3.68).
It should be clear that not every value of A and ω 2 gives a solution of (3.68). We will solve
for the allowed values by first finding the possible values of ω 2 and then finding the corre-
sponding values of A. To find the eigenvalues, note that (3.68) can be rewritten as
h i
M −1 K − ω 2 I A = 0 , (3.72)

where I is the n×n identity matrix. (3.72) is just a compact way of representing n homoge-
neous linear equations in the n components of A where the coefficients depend on ω 2 . We
saw in (3.47) and (3.48) that for systems of n homogeneous linear equations in n unknowns,
a nonzero solution exists if and only if the determinant of the coefficient matrix vanishes.
The reason is that if the determinant were nonzero, then the matrix, M −1 K − ω 2 I, would
have an inverse, and we could use (3.31) to conclude that the only solution for the vector, A,
is A = 0. Thus to have a nonzero amplitude, A, we must have
h i
det M −1 K − ω 2 I = 0 . (3.73)

(3.73) is a polynomial equation for ω 2 . It is an equation of degree n in ω 2 , because the term in

the determinant from the product of all the diagonal elements of the matrix contains a piece
£ 2 ¤n
that goes as ω . All the coefficients in the polynomial are real. Physically, we expect
all the solutions for ω 2 to be real and positive whenever the system is in stable equilibrium
because we expect such systems to oscillate. Mathematically, we can show that ω 2 is always
real, so long as all the masses are positive. We will do this below in (3.127)-(3.130).
Negative ω 2 are associated with unstable equilibrium. For example, consider a mass at
the end of a rigid rod, free to swing in the earth’s gravitational field in a vertical plane around
a frictionless pivot, as shown in figure 3.5. The mass can move along the dotted line. The
stable equilibrium position is indicated by the solid line. The unstable equilibrium position
is indicated by the dashed line.
When the mass is at the unstable equilibrium point, the smallest disturbance will cause it
to fall. Once away from equilibrium, the displacement increases exponentially until the angle
from the vertical becomes so large that the nonlinearities in the equation of motion for this
system take over. We will discuss this nonlinear oscillator further in appendix B.
Once we have found the possible values of ω 2 , we can put each one back into (3.72)
to get the corresponding A. Because (3.72) is homogeneous, the overall scale of A is not
determined, but all the ratios, aj /ak , are fixed for each ω 2 .

3.3.1 Normal Modes and Frequencies

The vector A is called the “normal mode” of the system associated with the frequency
ω. Because A is real, in the absence of friction, the complex solutions, (3.66), can be put
3.3. NORMAL MODES 71
.
.. .....
. . ... . . . . . . . . .
.....
......... ...
.. ... ...
.. ..
. ..
. ..
. .. .
. ..
. . .
. ..
..
. .. . ..
. .
.
. ..
. .
. ..
. ..
. ..
. . .
. . .
. .
.
. ... .
.
. . .
. . .
. . .
. .
. ..
. .
. .
. .
. .
. .. .
. . .
. . .
. .
. .... .
.
. . .
. .
. . .
...
.
.
.
.
...
.
.
• ..
.
.
. ..... .
. .
... .
. .
. ... .
. ... .
. ... .
. ... .
. ... .
. .
. .... .
. .
. ... .
. . .
. ... .
. .
.
.. .... .
.
.. ..... ..
.. .
... .
..
.. .... ..
.. .... ..
.. ... . .
.. ... ..
.. ... ..
.. .. ..
... . ..
... . . . .
.... ........ .
. . . ....... . ....... . . . . . . .
...... ......
..

Figure 3.5: A mass on a rigid rod, free to swing in the earth’s gravity in a vertical plane.

together into real solutions, like (3.69). The general real solution is of the form

X(t) = Re [(b + ic)Z(t)] =

(3.74)
b A cos ωt + c A sin ωt = d A cos(ωt − θ)

where b and c (or d and θ) are real numbers.

We can now construct the complete solution to the equation of motion. Because of lin-
earity, we get it by adding together all the normal mode solutions with arbitrary coefficients
that must be set by the initial conditions.
We can now see that the number of different normal modes is always equal to n, the
number of degrees of freedom. Label the normal modes as Aα , where α is a label that (we
will argue below) goes from 1 to n. Label the corresponding frequencies ωα . Then the most
general possible motion of the system is a sum of all the normal modes,
n
X
Z(t) = wα Aα e−iωα t (3.75)
α=1
72 CHAPTER 3. NORMAL MODES

or in real form (with w = b + ic)

n
X
X(t) = [bα Aα cos(ωα t) + cα Aα sin(ωα t)]
α=1 (3.76)
n
X
α
= dα A cos(ωα t − θα )
α=1

where bα and cα (or dα and θα ) are real numbers that must be determined from the initial
conditions of the system. Note that the set of all the normal mode vectors must be “com-
plete,” in the mathematical sense that any possible configuration of this system can be
described as a linear combination of normal modes. Otherwise, we could not satisfy arbi-
trary initial conditions with the solution, (3.76). This can be proved mathematically (because
the matrix, K, is symmetric and the masses are positive), but the physical argument will
be enough for us here. Likewise no normal mode can possibly be a linear combination of
the other normal modes, because each corresponds to an independent possible motion of the
physical system with its own frequency. The mathematical way of saying this is that the set
of all the normal modes is “linearly independent.”
Because the set of normal modes must be both complete and linearly in-
dependent, there must be precisely n normal modes, where again, n is the (3.77)
number of degrees of freedom.
If there were fewer than n normal modes, they could not possibly describe all possible
configurations of the n degrees of freedom. If there were more than n, they could not be
linearly independent n dimensional vectors. At least one of them could be written as a linear
combination of the others. As we will see later, (3.77) is the physical principle behind Fourier
analysis.
It is worth noting that solving the eigenvalue equation, (3.68), gets hard very rapidly as
the number of degrees of freedom increases. First you have to compute the determinant of
an n × n matrix. If all the entries are nonzero, this requires adding up n! terms. Once you
have finished that, you still have to solve a polynomial equation of degree n. For n > 3, this
cannot be done analytically except in special cases.
On the other hand, it is always straightforward to check whether a given vector is an
eigenvector of a given matrix and, if so, to compute the eigenvalue. We will use this fact in
the problems at the end of the chapter.

3.3.2 Back to the 2×2 Example

Let us return to the example from the beginning of this chapter in the special case where the
two pendulum blocks have the same mass, m1 = m2 = m. Simple as it is, this will be a very
important system for our understanding of wave phenomena. Let us see how the techniques
3.3. NORMAL MODES 73

that we have developed allow us to solve for the allowed frequencies and the corresponding
A vectors, the normal modes. From (3.7) and (3.8), the K matrix has the form
µ ¶
mg/` + κ −κ
K= . (3.78)
−κ mg/` + κ
The M matrix is µ ¶
m 0
M= . (3.79)
0 m
Thus from (3.78) and (3.79),
µ ¶
g/` + κ/m −κ/m
M −1 K = . (3.80)
−κ/m g/` + κ/m

The matrix M −1 K − ω 2 I is
µ ¶
g/` + κ/m − ω 2 −κ/m
M −1 K − ω 2 I = . (3.81)
−κ/m g/` + κ/m − ω 2

To find the eigenvalues of M −1 K, we form the determinant

·µ ¶¸
−1 2 g/` + κ/m − ω 2 −κ/m
det[M K − ω I] = det
−κ/m g/` + κ/m − ω 2

= (g/` + κ/m − ω 2 )2 − (κ/m)2 (3.82)

= (ω 2 − g/`)(ω 2 − g/` − 2κ/m) = 0 .

Thus the angular frequencies of the normal modes are

ω12 = g/` , ω22 = g/` + 2κ/m . (3.83)

To find the corresponding normal modes, we substitute these frequencies back into the
eigenvalue equation. For ω12 , the normal mode vector, A1 ,
µ 1¶
a
A1 = 1
, (3.84)
a12
satisfies the matrix equation
[M −1 K − ω12 I]A1 = 0 . (3.85)
From (3.81) and (3.83),
µ ¶
κ/m −κ/m
M −1 K − ω12 I = . (3.86)
−κ/m κ/m
74 CHAPTER 3. NORMAL MODES

Thus (3.85) becomes

µ ¶µ 1¶
κ/m −κ/m a 1
=0
−κ/m κ/m a12 (3.87)
µ ¶
κ a11 − a12
= ⇒ a11 = a12 .
m −a11 + a12
We can take a11 = 1 because we can multiply the normal mode vector by any number we like.
Only the ratio a11 /a12 matters. So, for example, we can take
µ ¶
1
A1 = . (3.88)
1

This gives (3.10). The displacement in this normal mode is shown in figure 3.6.

¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡
¡ ¡ ¡..... ¡ ¡ ¡ ¡ ¡..... ¡ ¡
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
.. ... .. . . .. ..
..... .... ...... .... ........... ........... .......... .......... .......... ..... ....
....... ....... ....... ...... ........ ........ ........ ........
................................................................................................
1 2

Figure 3.6: The displacement in the normal mode, A1 .

For ω22 , the normal mode vector, A2 ,

µ 2¶
a
A2 = 1
, (3.89)
a22

satisfies the matrix equation (where the identity matrix multiplying ω22 is understood)3

[M −1 K − ω22 ]A2 = 0 . (3.90)

3
It is tiresome writing the identity matrix, I, everywhere. It is not really necessary because you can always
tell from the context whether it belongs there or not. From now on, we will often leave it out. Thus, if you see
something that looks like a number in a matrix equation, like the −ω22 in (3.90), you should mentally include a
factor of I.
3.3. NORMAL MODES 75

This time, (3.81) and (3.83) give

µ ¶
−κ/m −κ/m
M −1 K − ω22 = . (3.91)
−κ/m −κ/m
Thus (3.90) becomes
µ ¶µ 2¶
−κ/m −κ/m a 1
=0
−κ/m −κ/m a22
µ ¶ (3.92)
κ a21 + a22
=− ⇒ a12 = −a22 .
m a21 + a22
Again, only the ratio a21 /a22 matters, so we can take
µ ¶
2 1
A = . (3.93)
−1
This gives (3.11). The displacement in this normal mode is shown in figure 3.7.

¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡
¡ ¡ ¡.... ¡ ¡ ¡ ¡ ¡..... ¡ ¡
. ...
... ...
... ...
... ...
.... ...
...
..
... ...
...
... ...
.... ...
.. ...
... ...
... ...
.... ...
...
.. ...
... ...
... ...
.... ...
.. ...
.. ...
.. ........ ........ ........ ....... ....... ....... .......
..... ..... .........
.... ... ... ..... .... .
...................................................................................................................................
1 2

Figure 3.7: The displacement in the normal mode, A2 .

The physics of these modes is easy to understand. In mode 1, the blocks move together
and the spring is never stretched from its equilibrium position. Thus the frequency is just g/`,
the same as an uncoupled pendulum. In mode 2, the blocks are moving in opposite directions,
so the spring is stretched by twice the displacement of each block. Thus there is an additional
restoring force of 2κ, and the square of the angular frequency is correspondingly larger.

3.3.3 n=2 — the General Case

Let us work out explicitly the case of n = 2 for an arbitrary K matrix,
µ ¶
K11 /m1 K12 /m1
M −1 K = , (3.94)
K12 /m2 K22 /m2
76 CHAPTER 3. NORMAL MODES

where we have used K21 = K12 . Then (3.73) becomes

à ! µ ¶
2
K11 K22 − K12 K11 K22
− + ω2 + ω4 = 0 , (3.95)
m1 m2 m1 m2

with solutions
s
µ ¶ µ ¶2 2
2 1 K11 K22 1 K11 K22 K12
ω = + ± − + . (3.96)
2 m1 m2 4 m1 m2 m1 m2
For each ω 2 , we can take a1 = 1. Then
m1 ω 2 − K11
a2 = . (3.97)
K12
As we anticipated, the eigenvectors turned out to be real. This a general consequence of
the reality of M −1 K and ω 2 . The argument is worth repeating. When all the elements of
the matrix M −1 K − ω 2 I are real, the ratios, aj /ak are real (because they are obtained by
solving a set of simultaneous linear equations with real coefficients). Thus if we choose one
component of the vector A to be real (multiplying, if necessary, by a complex number), then
all the components will be real. Physically, this means that for the solution, (3.66), all the
different parts of the system are oscillating not only with the same frequency, but with the
same phase up to a sign. This is true only because we have ignored damping. We will return
to the question in the last section (an optional section that is not for the fainthearted).

3.3.4 The Initial Value Problem

Once you have solved for the normal modes and corresponding frequencies, it is straightfor-
ward to put them together into the most general solution to the equations of motion for the
set of N coupled oscillators, (3.76). It is
X
X(t) = (bα Aα cos ωα t + cα Aα sin ωα t) . (3.98)
α

The 2N constants bα and cα are determined by the initial conditions. The bα are related to
the initial displacements, X(0):
X
X(0) = bα Aα . (3.99)
α

In words, bα is the coefficient of the normal¯ mode Aα in the initial displacement X(0). The
¯
cα are related to the initial velocities, dX(t)
dt ¯ :
t=0
¯ X
dX(t) ¯¯
= cα ωα Aα . (3.100)
dt ¯ t=0 α
3.4. * NORMAL COORDINATES AND INITIAL VALUES 77

The equations, (3.99) and (3.100), are two sets of simultaneous linear equations for the bα
and cα . They can be solved by hand. This is easy enough for a small number of degrees of
freedom. We will see in the next section that we can also get the solutions directly with very
little additional work by manipulating the normal modes.
Meanwhile, we should pause again to consider the physics of (3.98). This shows ex-
plicitly how the most general motion of the system can be decomposed into the simple mo-
tions associated with the normal modes. It is worth staring at an example (real, animated or
preferably both) at this point. Try to construct the system in figure 3.1. Any two identical
oscillators with a relatively weak spring connecting them will do. Convince yourself that the
normal modes exist. If you start the system oscillating with the blocks moving the same way
with the same amplitude, they will stay that way. If you get them started moving in opposite
directions with the same amplitude, they will continue doing that. Now set up a random mo-
tion. See if you can understand how to take it apart into normal modes. It may help to stare
again at program 3-1 on the program disk, in which this is done explicitly. In this animation,
you see the two blocks of figure 3.1 and below, the two normal modes that must be added to
produce the full solution.

3.4 * Normal Coordinates and Initial Values

There is another way of looking at the solutions of (3.14). We can find linear combinations
of the original coordinates that oscillate only with a single frequency, no matter what else is
going on. This construction is also useful. It allows us to use the form of the normal modes
to simplify the solution to the initial value problem.
To see how this works, let us return to the simple example of two identical pendulums,
(3.78)-(3.93). The most general possible motion of this system looks like
X(t) = bA1 cos(ω1 t − θ1 ) + cA2 cos(ω2 t − θ2 ) , (3.101)
or, using (3.88) and (3.93)

x1 (t) = b cos(ω1 t − θ1 ) + c cos(ω2 t − θ2 ) ,

(3.102)
x2 (t) = b cos(ω1 t − θ1 ) − c cos(ω2 t − θ2 ) .

The motion of each block is nonharmonic, involving two different frequencies and four con-
stants that must be determined by solving the initial value problem for both blocks.
But consider the linear combination
X 1 (t) ≡ x1 (t) + x2 (t) . (3.103)
In this combination, all dependence on c and θ2 goes away,
X 1 (t) = 2b cos(ω1 t − θ1 ) . (3.104)
78 CHAPTER 3. NORMAL MODES

This combination oscillates with the single frequency, ω1 , and depends on only two constants,
b and θ1 , no matter what the initial conditions are. Likewise,
X 2 (t) ≡ x1 (t) − x2 (t) (3.105)
oscillates with the frequency, ω2 ,
X 2 (t) = 2c cos(ω2 t − θ2 ) . (3.106)
X 1 and X 2 are called “normal coordinates.” We can just as well describe the motion
of the system in terms of X 1 and X 2 as in terms of x1 and x2 . We can go back and forth
using the definitions, (3.103) and (3.105). While x1 and x2 are more natural from the point
of view of the physical setup of the system, figure 3.1, X 1 and X 2 are more convenient for
understanding the solution. As we will see below, by going back and forth from physical
coordinates to normal coordinates, we can simplify the analysis of the initial value problem.
It turns out that it is possible to construct normal coordinates for any system of normal
modes. Consider a normal mode Aα corresponding to a frequency ωα . Construct the row
vector
B α = AαT M (3.107)
where AαT is the transpose of Aα , a row vector with aαj in the jth column.
The row vector B α is also an eigenvector of the matrix M −1 K, but this time from the
left. That is
B α M −1 K = ωα2 B α . (3.108)
To derive (3.108), note that (3.68) can be transposed to give
AαT K M −1 = ωα2 AαT (3.109)
because M −1 and K are both symmetric (see (3.18) and notice that the order of M −1 and K
are reversed by the transposition). Then
B α M −1 K = AαT M M −1 K = AαT KM −1 M (3.110)
= ωα2 AαT M = ωα2 B α . (3.111)
Given a row vector satisfying (3.108), we can form the linear combination of coordinates
X
X α = Bα · X = bαj xj . (3.112)
j

Then X α is the normal coordinate that oscillates with angular frequency ωα because
d2 X α α d X
2
= B · = −B α M −1 KX = −ωα2 B α · X = −ωα2 X α . (3.113)
dt2 dt2
Thus each normal coordinate behaves just like the coordinate in a system with only one
degree of freedom. The B α vectors from which the normal coordinates are constructed
carry the same amount of information as the normal modes. Indeed, we can go back
and forth using (3.107).
3.4. * NORMAL COORDINATES AND INITIAL VALUES 79

3.4.1 More on the Initial Value Problem

Here we show how to use normal modes and normal coordinates to simplify the solution of
the initial value problem for systems of coupled oscillators. At the same time, we can use
our physical insight to learn something about the mathematics of the eigenvalue problem. We
would like to find the constants bα and cα determined by (3.99) and (3.100) without actually
solving these linear equations. Indeed there is an easy way. We can make use of the special
properties of the normal coordinates. Consider the combination

B β Aα . (3.114)

This combination is just a number, because it is a row vector times a column vector on the
right. We know, from (3.112), that X β = B β X is the normal coordinate that oscillates with
frequency ωβ , that is:
B β X(t) ∝ e±iωβ t . (3.115)
On the other hand, the only terms in (3.98) that oscillate with this frequency are those for
which ωα = ωβ . Thus if ωβ is not equal to ωα , then B β Aα must vanish to give consistency
with (3.115).
If the system has two or more normal modes with different A vectors, but the same
frequency, we cannot use (3.115) to distinguish them. In this situation, we say that the modes
are “degenerate.” Suppose that A1 and A2 are two different modes with the same frequency,

M −1 K A1 = ω 2 A1 , M −1 K A2 = ω 2 A2 . (3.116)

Because the eigenvalues are the same, any linear combination of the two mode vectors is still
a normal mode with the same frequency,
³ ´ ³ ´
M −1 K β1 A1 + β2 A2 = ω 2 β1 A1 + β2 A2 , (3.117)

for any constants, β1 and β2 .

Now if A1T M A2 6= 0, we can use (3.117) to choose a new A2 as follows:

A1T M A2 1
A2 → A2 − A . (3.118)
A1T M A1
This new normal mode satisfies
A1T M A2 = 0 . (3.119)
The construction in (3.118) can be extended to any number of normal modes of the same
frequency. Thus even if we have several normal modes with the same frequency, we can still
use the linearity of the system to choose the normal modes to satisfy
T
B β Aα = Aβ M Aα = 0 for β =
6 α. (3.120)
80 CHAPTER 3. NORMAL MODES

We will almost always assume that we have done this.

We can use (3.120) to simplify the initial value problem. Consider (3.99). If we multiply
this vector equation on both sides by the row vector B β , we get
X X
B β X(0) = B β bα Aα = bα B β Aα = bβ B β Aβ . (3.121)
α α

where the last step follows because of (3.120), which implies that the sum over α only con-
tributes for α = β. Thus we can calculate bα directly from the normal modes and X(0),

B α X(0)
bα = . (3.122)
B α Aα
Similarly ¯
1 dX(t) ¯¯
ωα cα = Bα . (3.123)
α
B A α dt ¯t=0
The point is that we have already solved simultaneous linear equations like (3.99) in
finding the eigenvectors of M −1 K so it is not necessary to do it again in solving for bα
and cα . Physically, we know that the normal coordinate X α must be proportional to the
coefficient of the normal mode Aα in the motion. The precise statement of this is (3.122).

3.4.2 * Matrices from Vectors

We can also use (3.120) and the physical requirement of linear independence of the normal
modes to write M −1 K and the identity matrix in terms of the normal modes.
First consider the identity matrix. One can think of the identity matrix as a machine that
takes any vector and returns the same vector. But, using (3.120), we can construct such a
machine out of the normal modes. Consider the matrix H, defined as follows:
X Aα B α
H= . (3.124)
α B α Aα

Note that H is a matrix because Aα B α in the numerator is the product of a column vector
times a row vector on the right, rather than on the left. If we let H act on one of the normal
mode vectors Aβ , and use (3.120), it is easy to see that only the term α = β in the sum
contributes and H · Aβ = Aβ . But because the normal modes are a complete set of N
linearly independent vectors, that implies that H · V = V for any vector, V . Thus H is the
identity matrix,
H =I. (3.125)
We can use this form for I to get an expression for M −1 K in terms of a sum over
normal modes. Consider the product M −1 K · H = M −1 K, and use the eigenvalue condition
3.4. * NORMAL COORDINATES AND INITIAL VALUES 81

M −1 KAα = ωα2 Aα to obtain

X ω 2 Aα B α
α
M −1 K = . (3.126)
α B α Aα

In mathematical language, what is going on in (3.124) and (3.126) is a change of the basis
in which we describe the matrices acting on our vector space from the original basis of some
obvious set of independent displacements of the degrees of freedom to the less obvious but
more useful basis of the normal modes.

3.4.3 * ω 2 is Real
We can use (3.120) to show that all the eigenvalues of the M −1 K are real. This is a particular
example of an important general mathematical theorem. You will use it frequently when you
study quantum mechanics. To prove it, let us assume the contrary and derive a contradiction.
∗
If ω 2 is a complex eigenvalue with eigenvector, A, then the complex conjugate, ω 2 , is also
an eigenvalue with eigenvector, A∗ . This must be so because the M −1 K matrix is real, which
implies that we can take the complex conjugate of the eigenvalue equation,

M −1 K A = ω 2 A , (3.127)

to obtain
∗
M −1 K A∗ = ω 2 A∗ . (3.128)
∗
Then if ω 2 is complex, ω 2 and ω 2 are different and (3.120) implies

A∗ T M A = 0 . (3.129)

But (3.129) is impossible unless A = 0 or at least one of the masses in M is negative. To see
this, let us expand it in the components of A.
n
X n
X
A∗ T M A = a∗j mj aj = mj |aj |2 . (3.130)
j=1 j=1

Each of the terms in (3.130) is positive or zero. Thus the only solutions of the eigenvalue
equation, (3.127), for complex ω 2 are the trivial ones in which A = 0 on both sides. All the
normal modes have real ω 2 .
Thus there are only three possibilities. ω 2 > 0 corresponds to stable equilibrium and
harmonic oscillation. ω 2 < 0, in which case ω is pure imaginary, occurs when the equilibrium
is unstable. ω 2 = 0 is the situation in which the equilibrium is neutral and we can deform the
system with no restoring force.
82 CHAPTER 3. NORMAL MODES

3.5 * Forced Oscillations and Resonance

One of the advantages of the matrix formalism that we have introduced is that in matrix
language we can take over the above discussion of forced oscillation and resonance in chapter
2 almost unchanged to systems with more than one degree of freedom. We simply have to
replace numbers by appropriate vectors and matrices. In particular, the force F (t) in the
equation of motion, (2.2), becomes a vector that describes the force on each of the degrees
of freedom in the system. The only restriction here is that the frequency of oscillation is the
same for each component of the force. The ω02 in the equation of motion, (2.2), becomes
the matrix M −1 K. The frictional term ¡ becomes a matrix. In terms of the matrix ¡, the
frictional force vector is M ¡dZ/dt (compare (2.1)). Then we can look for an irreducible,
steady state solution to the equation of motion of the form

Z(t) = W e−iωt (3.131)

where W is a constant vector, which yields the matrix equation

h i
−ω 2 − i¡ω + M −1 K W = M −1 F0 . (3.132)

Formally, we can solve this by multiplying by the inverse matrix

h i−1
W = M −1 K − ω 2 − i¡ω M −1 F0 . (3.133)

If ¡ were zero in the matrix h i

−ω 2 − i¡ω + M −1 K , (3.134)
then we know that the inverse matrix would not exist for any value of ω corresponding to
a free oscillation frequency of the system, ω0 , because the determinant of the M −1 K − ω02
matrix is zero. The amplitude W would go to ∞ in this limit, in the direction of the normal
mode associated with the driving frequency, so long as the driving force has a component
in the normal mode direction. For ω close to ω0 , if there is no damping, the response
amplitude is very large, proportional to 1/(ω02 − ω 2 ), almost in the direction of the
normal mode. However, in the presence of damping, the response amplitude does not go to
∞ even for ω = ω0 , because the i¡ω term is still nonvanishing.
We can see all this explicitly if the damping matrix ¡ is proportional to the identity matrix,

¡=γI. (3.135)
£ ¤
Then we can use (3.124)-(3.126) to write M −1 K − ω 2 − i¡ω as a sum over the normal
modes, as follows:
h i X³ ´ Aα B α
M −1 K − ω 2 − i¡ω = ωα2 − ω 2 − iγω . (3.136)
α B α Aα
3.5. * FORCED OSCILLATIONS AND RESONANCE 83

Then the inverse matrix can be constructed in a similar way, just by inverting the factor in the
numerator:
h i−1 X³ ´−1 Aα B α
M −1 K − ω 2 − i¡ω = ωα2 − ω 2 − iγω . (3.137)
α B α Aα

Using (3.137), we can rewrite (3.133) as

X Aα B α M −1 F0
W = . (3.138)
α ωα2 − ω 2 − iγω B α Aα

This has a simple interpretation. The second factor on the right hand side of (3.138) is the
coefficient of the normal mode Aα in the driving term, M −1 F0 . This coefficient is multiplied
by the complex number · ¸
1
, (3.139)
ωα2 − ω 2 − iγω
which is exactly analogous to the factor in (2.21) in the one dimensional case. Thus if ¡ ∝ I,
then, for each normal mode, the forced oscillation works just as it does for one degree of free-
dom. If ¡ is not proportional to the identity matrix, the formulas are a bit more complicated,
but the physics is qualitatively the same.

3.5.1 Example
We will illustrate these considerations with our favorite example, the system of two identical
coupled oscillators, with M −1 K matrix given by (3.80). We will imagine that the system is
sitting in a viscous fluid that gives a uniform damping ¡ = γI, and that there is a periodic
force that acts twice as strongly on block 1 as on block 2 (for example, we might give the
blocks electric charge 2q and q and subject them to a periodic electric field), so that the force
is µ ¶ ·µ ¶ ¸
2 2
F (t) = f0 cos ωt = Re f0 e−iωt . (3.140)
1 1
Thus µ ¶
2 f0
M −1 F0 = . (3.141)
1 m
Now to use (3.133), we need only invert the matrix
µg ¶
−1 2 + κ − ω 2 − iγω −mκ
[M K − ω − i¡ω] = ` m κ g κ . (3.142)
− m ` + m − ω 2 − iγω
84 CHAPTER 3. NORMAL MODES

This is simple enough to do by hand. We will do that first, and then compare the result with
(3.137). The determinant is
µ 2 ¶ µ ¶
g κ κ 2
+ − ω 2 − iγω −
` m m (3.143)
µ ¶ µ ¶
g κ g
= + 2 − ω 2 − iγω · − ω 2 − iγω .
` m `
Applying (3.34), we find

[M −1 K − ω 2 − i¡ω]−1

1 (3.144)
= ¡g ¢¡ ¢
` + κ
2m − ω2 − iγω g` − ω 2 − iγω
µg κ 2 κ ¶
· ` + m − ω − iγω m .
κ g κ
m ` + m − ω 2 − iγω

If we isolate the contribution of the two zeros in the denominator of (3.144), we can write

[M −1 K − ω 2 − i¡ω]−1

µ ¶
1 1 1 1 (3.145)
= ¡g ¢
2
2 ` − ω − iγω 1 1
µ ¶
1 1 1 −1
+ ¡g κ
¢
2 ` + 2 m − ω 2 − iγω −1 1
3.5. * FORCED OSCILLATIONS AND RESONANCE 85

which is just (3.137), as promised. Now substituting into (3.133), we find

µ ¶
1 1 3 f0
W = ¡g ¢
2
2 ` − ω − iγω 3 m
µ ¶
1 1 1 f0
+ ¡g κ
¢
2 ` + 2 m − ω 2 − iγω −1 m
(3.146)
¡g ¢ µ ¶
1 `− + iγωω2 3 f0
= ¡g ¢2
2 − ω 2 + (γω)2 3 m
`
¡ ¢ µ ¶
1 g` + 2 m
κ
− ω 2 + iγω 1 f0
+ ¡g ¢ ,
2 + 2 κ − ω 2 2 + (γω)2 −1 m
` m

from which we can read off the final result:

³ ´ µ ¶
α1 cos ωt + β1 sin ωt
X(t) = Re W e−iωt = (3.147)
α2 cos ωt + β2 sin ωt

where
g 2
¡ ¢
3 ` −ω f0
α1(2) = ¡g ¢2 2
2 − ω 2 + (γω) m
` (3.148)
¡g κ 2
¢
1 ` + 2m − ω f0
± ¡g ¢2
2 +
` 2mκ
− ω 2 + (γω)2 m
and

3 γω f0
β1(2) = ¡g ¢2 2
2 −ω 2 + (γω) m
` (3.149)
1 γω f0
± ¡g ¢ 2 2
.
2 + 2 κ − ω 2 + (γω) m
` m
The power expended by the external force is the sum over all the degrees of freedom of
the force times the velocity. In matrix language, this can be written as

dX(t)
P (t) = F (t)T · . (3.150)
dt
86 CHAPTER 3. NORMAL MODES

The average power lost to the frictional force comes from the cos2 ωt term in (3.150) and is

1 9γω 2 f02
= ¡g ¢2
` − ω2 + (γω)2 4m
(3.151)
1 γω 2 f02
+¡g ¢2
`
κ
+ 2m − ω2 + (γω)2 4m

Figure 3.8 shows a graph of this (for κ/m = 3g/2` and γ 2 = g/4`). There are two
things to observe about figure 3.8. First note the two resonance peaks, at ω 2 = g/` and
ω 2 = g/` + 2κ/m = 4g/`. Secondly, note that the first peak is much more pronounced that
the second. That is because the force is more in the direction of the normal mode with the
lower frequency, thus it is more efficient in exciting this mode.
.....
.. ....
... ...
.... ....
.. ...
... ....
... ...
.... ...
...
.. ...
.... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
...
... ...
... ...
... ...
... ...
... ...
...
... ...
... ...
... ...
... ....
... .....
........................
.
. .....
... .....
.....
... ......
..
. .......
..........
.
.. ................
. .
... ......
..
.
......
.
0 .
.........
...
p p
0 g/` 2 g/`
ω→

Figure 3.8: The average power lost to friction in the example of 3.140.

Chapter Checklist
You should now be able to:

1. Write down the equations of motion for a system with more than one degree of freedom
in matrix form;

2. Find the M and K matrices from the physics;

3. Add, subtract and multiply matrices;

PROBLEMS 87

4. Find the determinant and inverse of 2×2 and 3×3 matrices;

5. Find normal modes and corresponding frequencies of a system with two degrees of
freedom, which means finding the eigenvectors and eigenvalues of a 2×2 matrix;

6. Check whether a given vector is a normal mode of a system with more than two degrees
of freedom, and if so, find the corresponding angular frequency;

7. Given the normal modes and corresponding frequencies and the initial positions and
velocities of all the parts in any system, find the motion of all the parts at all subsequent
times;

8. * Go back and forth from normal modes to normal coordinates;

9. * Reconstruct the M −1 K matrix from the normal modes and normal coordinates;

10. * Explicitly solve for the free oscillations of system with two degrees of freedom with
damping and be able to analyze systems with three or more degrees of freedom if you
are given the eigenvectors;

11. * Explicitly solve forced oscillation problems with or without damping for systems
with three or fewer degrees of freedom.

Problems

3.1. The 3 component column vector A, the 3 component row vector B and the 3 × 3
matrix C are defined as follows:
⎛ ⎞ ⎛ ⎞
0 1 1 1
A = ⎝2⎠ , B = ( 3 −2 1 ) , C = ⎝0 −2 1 ⎠ .
1 2 2 0

Compute the following objects:

BA , BC , AB .

3.2. Consider the vertical oscillation of the system of springs and masses shown below
with the spring constants KA = 78, KB = 15 and KC = 6 (all dynes/cm). Find the normal
modes, normal coordinates and associated angular frequencies. If the 1 g. block is displaced
88 CHAPTER 3. NORMAL MODES

up 1 cm from its equilibrium position with the 3 g block held at its equilibrium position and
both blocks released from rest, describe the subsequent motion of both blocks.
... . ..
................
. ......... .........................
.. ..
..... . ..
.............. ......................
.............. ...... ..
... ........ .. .. ..
.
.
............... ....................
..... ...
KA ...............
.
.. ..
......................
.
..... ......... ...... ..
.
.............. .. .. . .
.
.............. ...................
.. ..
... .
......................
...... ..
3 g. .... . .
...................
.. ..
.
.. KC
... . .
...............
. .......... ..........................
.. ..
..... . ..
...............
. .......... .........................
.. ..
KB .... .
................
. ..........................
.
..... ......... .. ..
.
.............. .... . . ..
............... .................
.. ..
.... . .
................ ..........................
.
.... ......... .. ..

1 g.

3.3. Consider the system of springs and masses shown below:

2110 90 81 1701
.. .. . .. .. . .. .. . .. .. .
..... ..... ..... ..... ............ ..... ..... ...... ..... ............ ..... ..... ...... ..... ............ ..... ..... ...... ..... ............
.... .... ....
.......................................... 1 .... .... ....
.......................................... 2 .... .... ....
.......................................... 3 .... .... ....
..........................................

with the spring constants in newtons/meter given above the springs and with m1 = 100 kg,
m2 = 9 kg and m3 = 81 kg.

a. Which of the following are normal modes of the system and what are the correspond-
ing angular frequencies? Note that the M −1 K matrix may look a little complicated.
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
ψ1 9 9 9 9 9
⎝ ψ2 ⎠ = ⎝ 0 ⎠ ⎝ 60 ⎠ ⎝ −30 ⎠ ⎝ 30 ⎠ ⎝ 0 ⎠
ψ3 10 10 10 10 −10

b. If the system is released from rest with an initial displacement as shown below (with
the displacements measured in mm), how long does it take before it first returns to its initial
configuration? ⎛ ⎞ ⎛ ⎞
ψ1 9
⎝ ψ2 ⎠ = ⎝ 0 ⎠
ψ3 10
PROBLEMS 89

3.4 * . A system of four masses connected by springs is described by a mass matrix,

⎛ ⎞
1 0 0 0
⎜0 2 0 0⎟
M =⎜
⎝0
⎟
0 1 0⎠
0 0 0 2
and a K matrix ⎛ ⎞
29 −10 −4 −2
⎜ −10 58 −14 −2 ⎟
K=⎜
⎝ −4 −14
⎟
31 −26 ⎠
−2 −2 −26 74

a. Which of the following are normal modes?

⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 1 2 2 4 0
⎜2⎟ ⎜1⎟ ⎜1⎟ ⎜ 1 ⎟ ⎜ −3 ⎟ ⎜ 1 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎝1⎠ ⎝2⎠ ⎝1⎠ ⎝ −1 ⎠ ⎝ 0 ⎠ ⎝ −4 ⎠
1 1 1 −1 1 3

b. For each normal mode, find the corresponding angular frequency. Hint: this requires
a little arithmetic. If you are lazy, you might want to use a programmable calculator or write
a little computer program to check these for you. But the point of this problem is to show
you that the amount of work required to check whether the vectors are normal modes is really
tiny compared to the work involved in finding the modes from scratch.

c. If blocks are released from rest from an initial displacement that is proportional to
⎛ ⎞
1
⎜ 1 ⎟
⎜ ⎟
⎝ −1 ⎠ ,
1
which normal mode is not present in the subsequent motion?

d. Find the normal coordinates corresponding to each of the normal modes of the sys-
tem.

3.5. Consider the longitudinal oscillations of the system shown below:

15 dyne
cm 90 dyne
cm 10 dyne
cm

... .. .
..... ..... ...... ..... ............ ............ ........... ............ ............ ............ ........... ........... ............ ............ ........... ........... ........... ...........
.... .... .... ..... ... .. .. .. 1 ..... .. .. ... .. ...... 2 ..... ..... .. .. .. .. .. .. .. ..
.................................................................................... ........................................................ ..........................................................................................
90 CHAPTER 3. NORMAL MODES

The blocks are free to slide horizontally without friction. The displacements of the blocks
from equilibrium are both measured to the right. Block 1 has a mass of 15 grams and block
2 a mass of 10 grams. The spring constants of the springs are shown in dynes/cm.

a. Show that the M −1 K matrix of this system is

µ ¶
7 −6
M −1 K = .
−9 10

b. Show that the normal modes are

µ ¶ µ ¶
1 1 2 2
A = , A = .
1 −3
Find the corresponding angular frequencies, ω1 and ω2 .

3.6. Consider the longitudinal oscillations of the system shown below:

K1 K2 K3

... . .
..... .... ........... ............ ............ ... . .
..... ..... ............ ............ ........... ............ ............ ............ ............ ............ ............ ........... ...........
.... .... .... .....
........................................................ 1 .... .... .... ... .. .. .. .. .. .. .. .. ..
................................................................................................................ 2 ..... ..... ...... .. ..
..........................................................

The blocks are free to slide horizontally without friction. The displacements of the blocks
from equilibrium are both measured to the right. Block 1 has a mass of 15 grams and block 2
a mass of 10 grams. The spring constants of the springs are K1 , K2 and K3 , as shown. The
normal modes of this system are
µ ¶ µ ¶
1 2 2 1
A = , A =
3 −1
with corresponding frequencies

ω1 = 1 s−1 , ω2 = 2 s−1 .

a. If the system is at rest at time t = 0 with displacements x1 (0) = 5 cm, x2 (0) = 0, or

µ ¶ µ ¶
x1 (0) 5
X(0) = = cm .
x2 (0) 0
Find the displacement of block 2 at time t = π s.

b. Find K1 , K2 and K3 .
PROBLEMS 91

3.7 * . In the system of problem (3.5), suppose we immerse the system in a damping
fluid so that µ ¶
γ 0
¡=
0 γ
with γ = 1 s−1 , and that an external force of the following form is applied (in dynes):
µ ¶
1
F (t) = f cos ωt = cos ωt .
0

Find and graph the average power lost to the frictional force as a function of ω from ω = 0
to 10 s−1 .
MIT OpenCourseWare
https://ocw.mit.edu

8.03SC Physics III: Vibrations and Waves

Fall 2016

For information about citing these materials or our Terms of Use, visit: https://ocw.mit.edu/terms.